Voltage regulator devices typically maintain terminal voltages of voltage sources within required limits despite variations in input voltages or loads. Industry standard voltage regulators, also known as DC/DC converters, are generally not fault tolerant, thus the output of the regulator goes out of regulation during a fault. Most of the point-of-load power converters in server systems, like the voltage regulator modules (VRMs) for Intel processors, are of the same topology: non-isolated, step-down (buck) converters with synchronous rectification. A basic schematic diagram of a standard buck converter is illustrated in FIG. 1. The distinguishing feature of a synchronous buck converter is that a lower switch S2 is implemented by a diode 10 in parallel with a field effect transistor (FET) 12. In some designs, the body diode of the MOSFET is used for the function of a discrete diode 10. In this case an open circuit, say due to a wire bond failure would also be detected. Switch S2 and another switch S1, formed by a FET 14, are controlled in a complimentary fashion, such that either one or the other switch, but not both switches, is ON, except for a small `dead-time` when only the diode 10 conducts. Efficiency is achieved by the arrangement, because the losses of existing FET devices are generally better than those of existing diode devices. The arrangement further allows current to flow through S2 in reverse, and thus synchronous converters can regulate down to zero DC load.
Many converters, such as that in FIG. 1, also use a control scheme called `current-mode` control, where output current is sensed through a resistor 16 (R1), which is normally in the range of 3 to 10 milliohms. The resulting information is then used to help control the converter and the output voltage feedback via control unit 18. A clock signal, CLOCK, in the control unit 18 sets the switching frequency and is the basis of timing inside the converter. The base period of the switching, Tclock, results from the clock signal. A switching node, V1, acts as a summing point of switches S1, S2, and inductor L1. Under normal operations, the voltage at V1 has a rising edge synchronized, except for propagation delays, with the clock signal and a falling edge set by the control unit 18. Excluding losses, during DC conditions, the period that the signal at V1 is on, Ton, =Tclock (Vout/Vin). The rectangular waveform resulting at V1 is then chopped down by a filter formed by the inductor 19 (L1) and a capacitor 21 (C1). In order to keep the output voltage ripple at low levels required by a load device, e.g., a CPU (central processing unit), the corner frequency of the L1-C1 filter is virtually always kept at least 10 times lower in frequency than the frequency of the clock signal.
Faults in a voltage regulator can be problematic and are usually not detected until the regulator goes out of tolerance as detected by a fault detection device coupled to the output of the regulator. "Up-time" is becoming increasingly important in servers as the servers take on tasks once performed by ultra-reliable mainframes. Redundancy is the typical method used to achieve a high degree of basic reliability in servers. FIG. 2 illustrates a plurality of redundant voltage regulator modules 20, VRM1 to VRMn, which are coupled in parallel to a sensitive load device 22, e.g., a CPU, such that if one VRM 20 goes down, another VRM takes over. Without fault detection hooks to sense failures in redundant elements and a way in which to report them, however, redundancy is significantly less useful.
Isolating faults is also problematic. Prior art approaches typically use a diode on the output of the converter in order to provide a means to isolate parallel converters for fault tolerance reasons. Due to their finite voltage drop, "OR-ing" diodes to isolate a faulted converter significantly decreases the efficiency of the system. Further, the diodes make it nearly impossible to meet stringent dynamic load performance requirements of certain specifications, e.g., Intel specifications. Simply tieing the converters together offers one possible alternative approach, but the overall reliability actually decreases under such circumstances, since many faults in one converter may cause either an overcurrent or overvoltage condition, which brings down the entire system of parallel converters. While replacing the diode on the output with a semiconductor switch, such as a power MOSFET, has been attempted, the problems of the OR-ing of the diodes remain, albeit in a somewhat diminished capacity.
Accordingly, what is needed is a method and system for detecting faults in a redundant power converter, e.g., voltage regulator module, before the output voltage goes out of regulation.